The Technological Obsolescence of Mediocrity

NOT FOR QUOTATION WITHOUT PERMISSION OF THE AUTHOR

THE

Ronald M. Lee

August 1982 WP-82-76

Working Papers a r e interim reports on work of the International Institute for Applied Systems Analysis and have received only limited review. Views or opinions expressed herein do not necessarily represent those of the Institute or of its National Member Organizations.

INTERNATIONAL INSTITUTE FOR APPLIED SYSTEMS ANALYSIS 2361 Laxenburg, Austria

Will artificially intelligent expert systems necessarily serve to amplify human intelligence? Or will they simply create another wave of technological displacement? What types of occupations will be affected?

What effects will these technologies have on developing human expertise?

THE IXCHNOLOGICAL OBSOLESCENCE OF MEDIOCRITY

Ronald

M.

mediocrity ... the quality of being mediocre; spec. a moderate or average degree of mental ability, talents, skill, or the like; mid- dling capacity, endowment, or accomplishment"

-

In a n earlier issue, Joseph (1982) wrote of the possibilities of (human) intelligence amplification through the use of artificially intelli- gent machines. His article examined a variety of technological develop- ments which make this a possibility. However the question remains open as to how these developments will combine with human abilities to attain higher capacities of cognition.

Artificial Intelligence research has always emphasized the similari- ties between the observable evidence of human cognition and the behavior of computer programs. The comparison has been used in two ways: one, as a psychological methodology, using computer programs as a possible model of human cognition; the other as an engineering orienta- tion, using human cognition as a model for building smarter computer systems.

However, by accepting these similarities as the basis for combining computers and humans in a single category of 'cognitive entities,' we are likewise led to focus on their differences as well.

On one hand, there is a fairly well developed literature (e.g., Miller 1956, Tversky and Kahneman 1974, Simon 1981) which emphasizes the limitations of human cognition with respect to machines. These deal mainly with the limitations of short term memory, coupled with relatively slow sequenbal processing capability w h c h lead us (humans) to simplify problems by abstracting their components into larger 'chunks,' and using

short-cut heuristics to trim down the problem's complexity.

On the other hand, another literature is emerging (e.g., Weizenbaum 1976, H. Dreyfus 1979, S. Dreyfus 1981). which emphasizes the limita- tions of computational cognition as compared to that of humans. The general criticism is that computational techniques rely on atomistic representations of d a t a and the sequential application of separate and exact inference rules whereas human (organic) cognition appears to store wholistic impressions and images and is capable of fuzzy* p a t t e r n match- ing between t h e m which allows for great flexibility of association.

This suggests a theory of cognitive complementarity between human and machines.

Humans for instance require a great deal of discipline and training to perform the types of iterative calculations most easily programmed in machines.

Contrariwise, the types of cognition w h c h a r e basic even t o human infants

-

-

present deep, unsolved problems for computational theories.

The middle a r e a where humans and machines appear to be o n com- parable footing is in so-called 'rule-based' system which form the general architecture of expert systems applications (see Davis and King (1975) or Nilson (1 980) for background).

Rather than procedural programs where the computer executes instruction after instruction in a pre-determined order, these are non- procedural programs of un-ordered rules where the machine searches repeatedly through the rule set for the appropriate rules pertaining to a given situation. The non-determinism of this approach sacrifices much of t h e efficiency where the computer normally has advantage over t h e human. On the other hand, it provides considerably more flexibility and adaptability w h c h are the human's normal advantage.

S t u a r t Dreyfus (1981) makes some interesting observations regard- ing rule-based cognition in the formation of human expertise. His claim is that the use of a small s e t of discrete rules is characteristic of the novice stage in the development of a particular skill. As the individual becomes more experienced, these rules are gradually refined t o incor- porate numerous exceptions. Additional experience adds a context dependent organization to the rules as well a s additional refinement so t h a t the rules take on a much broader, parametric character. In the case of more advanced expertise, the individual rules give way t o more wholistic patterns w b c h are no longer processed in sequence but r a t h e r in a simultaneous pattern oriented manner. He suggests mundane exam- ples such as learning to drive a c a r or playing chess. The novice driver learns to s h f t a t specified velocities, has certain fixed procedures for parallel parking, e t c . Experienced drivers, on the other hand, no longer rely on these elementary measures but rather incorporate a wide variety of factors such as the sound of the engine, road incline and surface condi- tion, weather, anticipated traffic situations, e t c . A key point is that a t

*

this level, most experienced drivers can no longer specify the individual factors and rules they use.

Likewise most novice chess players begin with a simple point valua- tion scheme f o r each of the players and evaluate the value of an exchange through this numeric comparison. Subsequent development adds con- sideration of the relative position of pieces and their projected positions through scenarios of play and counter-play. Evidence of master level chess play however suggests a much more holistic orientation depending on comparative 'field of force' in actual and potential configurations of t h e pieces.

Along these lines, a dimly recalled anecdote tells of well-known and widely recognized designer of electric motors. His fame grew to the point where he decided to write a book explaining how he did his designing.

After the book was published, the quality of his subsequent designs declined. The conjecture is that he felt obligated t o follow the rules and methods of his book which somehow failed to capture the full capacity of his former expertise.

The general hypothesis here is that the major impact of rule-based, expert systems will be a t these types of cognition characteristics of the early to middle level stages of human expertise development.

These types of considerations from the basic design philosophy of so-called "decision support systems" e.g., Keen & Scott-Morton (1978),, Sprague & Fick (1980), Bonczek e t al. (1 981). As opposed to expert sys- tems, which attempt to replicate the abilities of a human expert in a specific problem domain, the aim behind decision support systems is to arrive a t a symbiotic combination between person and m a c h n e . These have been of especial interest in management applications where deci- sions are a t best only partially formalizable and continue to rely on experienced judgment. It is in these types of decision support applica- tions that the notation of intelligence amplification holds the most prom- ise.

On the other hand, whle these developments have a certain exotic fascination, they also suggest a new wave of technological displacement.

Rather than the typical victims such as clerks or factory workers, this wave threatens t o shake occupations of the middle level professional, e . g . , in banking, law, medicine, public administration.

A great deal of this intellectual activity is the application of esta- blished professional principles. While it often takes extensive specialized training to acquire these abilities, the very process of specialization tends to make these disciplines amenable to rule-based, automatable represen- tations. As observed in Lee (19B0), the Weberian concept of rationaliza- tion of organization has already become closely linked with automation.

Supposing these developments do occur, what would be the potential impact on expertise formation in these fields?

We would like to think that by removing the more mechanical types of cognition, human abilities will be freed to address problems requiring higher levels of creativity and innovation. A social question is whether their a r e enough such challenging problems to go around. The corresponding psychological question is whether the displaced

professionals will have the creative aptitude to address these challenges.

There is a pedagogic issue here as well. If rule based cognition is a n important initial stage in expertise acquisition, what will be the effect if this stage loses its rewards? Will this tend to block the development of further experts?

The example of chess may provide us the first experimental evi- dence. The more sophisticated compute chess programs now approach

"expert" (in chess terminology) levels of play. The most advanced "mas- ters" level chess players continue relatively unchallenged by these machines. Nonetheless it now takes a minimum of several years concen- trated chess study to beat a machine. Even the inexpensive commer- cially marketed chess machines are a match for intermediate level players. What effect will this have for potential chess students? Will they be challenged or merely discouraged?

What would be the effect if the problem domain were law or engineer- ing and the only jobs available are beyond the machine's capability? Will short t e r m intelligence amplification become long term intelligence stag- nation?

REFERENCES

Bonczek, R.H., C.W. Holsapple and A.B. Whnston. 1981. Foundations o f Decision Support S y s t e m s . New York: Academic Press.

Davis, R , and J. King. 1975. An Overview of Production Systems. Stanford A1 Lab Memo AIM-271, Stanford Computer Science Report. STAN-CS- 75-524.

Dreyfus, H. 1979. What Computers Can't Do. Revised Edition, New York:

Harper Colophon Books, Harper and Row.

Dreyfus, S.E. 1981. Formal Models vs. Human Situational Understanding:

Inherent limitations on the Modeling of Business Expertise. Berke- ley, CA: Operations Research Center, University of California, Berke- ley.

Joseph, E.C. 1982. What's Ahead for Intelligence Amplification? FUTUR- ICS 6(2):35-38.

Keen, P.G.W. and M.S. Scott-Morton. 1978. Decision Support S y s t e m s : An Organizational Perspective. Reading, MA: Addison-Wesley.

Lee, R.M. 1980. Bureaucracies, Bureaucrats and Technology. WP-80-186.

Laxenburg, Austria: International Institute for Applied Systems Analysis.

Miller, G.A. 1956. The Magical Number Seven, Plus or Minus Two: Some Limits on O u r Capacity for Processing Information. P s y c h o l o g i c a l R e v i e w 63(2):81-97.

Nilsson, N.J. 1980. P r i n c i p l e s of A r t i f i c i a l I n t e l l i g e n c e . Palo Alto, CA:

Tioga Publishing.

Simon, H.A. 1981. Studying Human Intelligence by Creating Artificial Intelligence. A m e r i c a n S c i e n t i s t 69(May-June):300-309.

Sprague R.H., and G. Fick. 1980. D e c i s i o n S u p p o r t S y s t e m s : I s s u e s a n d C h a l l e n g e s . Oxford: Pergamon Press.

Tversky A. and D. Kahneman. 1974. Judgment under Uncertainty: Heuris- tics and Biases. S c i e n c e 185:1124-1131.

Weizenbaum, J. 1976. C o m p u t e r P o w e r a n d H u m a n R e a s o n . San Fran- cisco: Freeman.

Zadeh, L.A., K.-S. Fu, K. Tanaka and M. Shimura (eds.). 1975. Fb~zzy S e t s a n d T h e i r A p p l i c a t i o n s t o C o g n i t i v e a n d Decision P r o c e s s e s . New York: Academic Press.